The System.Random is just not very good, you should not use it if you need
something fast or cryptographically secure... But as a first choice for
just a few random numbers it's good enough and it was chosen specifically
for its capability to split a generator into two, so that's probably what
you should be using if you're going to do simulations in parallel with
System.Random. Or use random seeds at least !
This is specifically an interaction between random*R* and mkStdGen with low
seeds (random don't seem to have this particular flaw), since we're still
using a pseudo-random generator, we ought to be pretty careful in the way
we use it (those have always been touchy beasts) and this particular usage
seems to be a bad counter-example of what to do with a PRNG.
If we can improve System.Random to avoid this particular misbehavior we
should but using a PRNG in this fashion and hoping for good randomness from
the result is probably a bad idea in the first place.
Le mer. 4 nov. 2015 à 11:00, Dominic Steinitz
http://article.gmane.org/gmane.comp.lang.haskell.beginners/15925
I do think this is a flaw and catches many people out despite apparently being well documented. And it's something one probably wants e.g. to run multiple Markov Chain Monte Carlo simulations in parallel.
Some further information here: https://github.com/haskell/random/issues/30#issuecomment-153647055
Dominic
_______________________________________________ Beginners mailing list Beginners@haskell.org http://mail.haskell.org/cgi-bin/mailman/listinfo/beginners